Fix incorrect Tensor Size for NestedTensor QKV Transform #2450
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Contributor
There was a problem hiding this comment.
Pull request overview
This PR fixes a tensor size mismatch issue in the transform_bias_rescale_qkv_xpu function when processing NestedTensors. The problem occurred when the sequence length (T) wasn't a multiple of 8 after implicit padding, causing the Q tensor output size to differ from the expected reference size.
Key Changes:
- Added explicit padding logic to round up the sequence length T to the next multiple of 8 for NestedTensor cases
- This ensures output tensor dimensions align with Tensor core requirements and match the Python reference implementation
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
CuiYifeng
force-pushed
the
yucai/mha/nested/fix
branch
from
590ebe6 to
980872b
Compare
CuiYifeng
changed the title
liangan1
reviewed
Dec 9, 2025
| // cores. Otherwise, sometimes with padding, *no* row will have the maximum | ||
| // sequence length and so we'll have a non-divisible-by-8 dimension even if | ||
| // the model author chose a multiple of 8. | ||
| T = T + (8 - (T % 8)) % 8; |
There was a problem hiding this comment.
The dpas should not have limitation, the m can be 1~8, why we need to change it here?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
To solve #2182 : Q (Query) tensor output size from torch.transform_bias_rescale_qkv was mismatched against the expected reference size in test cases involving Nested Tensors where the sequence length (T) was not a multiple of 8 after implicit padding.
Resolution: The resolution involved introducing logic within the C++ function transform_bias_rescale_qkv_xpu specifically for the Nested Tensor case to explicitly use the calculated sequence length T to resize the output q, k, and v tensors, thereby ensuring their final size matches the shape derived by the Python reference implementation.